Tether’s AI division QVAC has released Genesis III, a dataset of 191.43 billion tokens designed to train smaller AI models in STEM reasoning. The dataset covers 19 STEM domains and uses a dual teacher-distillation strategy: the system generates corrective explanations from a model’s errors and contrastive reasoning from its successes. Models trained on Genesis III achieved performance gains of up to 28.57% on ARC-Easy and 21.35% on ARC-Challenge compared to existing open datasets. The dataset is freely available on Hugging Face under Creative Commons CC-BY-NC 4.0 license. This release reflects Tether’s broader strategic push toward decentralized, on-device AI.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
